speakrs models

This repository stores the model artifacts used by speakrs, a fast Rust speaker diarization library.

With the default online feature, speakrs downloads the required files from this repository on first use. The SDK currently pins revision 5d24ffee75f13fb061fa6d10944a64e2dc1d5e6f.

Contents

• ONNX segmentation and embedding artifacts for CPU, CUDA, and MIGraphX runs
• CoreML .mlmodelc bundles for Apple-platform CoreML runs
• PLDA and VBx parameter files used by the clustering pipeline

Usage

# macOS with CoreML
speakrs = { version = "0.5", features = ["coreml"] }

# NVIDIA GPU
speakrs = { version = "0.5", features = ["cuda"] }

# CPU only
speakrs = "0.5"

use speakrs::{ExecutionMode, OwnedDiarizationPipeline};

fn main() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    let mut pipeline = OwnedDiarizationPipeline::from_pretrained(ExecutionMode::CoreMl)?;

    let audio: Vec<f32> = load_your_mono_16khz_audio_here();
    let result = pipeline.run(&audio)?;

    print!("{}", result.rttm("my-audio"));
    Ok(())
}

For offline or airgapped setups, download this repository and set SPEAKRS_MODELS_DIR to the local model directory.

Provenance

The artifacts are exported or converted for speakrs from the pyannote community-1 pipeline and its segmentation and WeSpeaker components. PLDA and VBx parameters are extracted from the pipeline cache. CoreML bundles are converted from the exported model artifacts for Apple-platform execution.

Upstream model access may require accepting upstream terms. Users are responsible for complying with the licenses and terms of the upstream models and datasets.

Links

• GitHub: https://github.com/avencera/speakrs
• Crates.io: https://crates.io/crates/speakrs
• Documentation: https://docs.rs/speakrs/latest/speakrs/